On Mon, Aug 27, 2012 at 11:56:32PM -0400, Tom Lane wrote:
> Bruce Momjian <bruce@momjian.us> writes:
> > On Fri, Mar 9, 2012 at 04:36:08PM -0800, David E. Wheeler wrote:
> >> A <type>bigint</type> key is displayed with its
> >> high-order half in the <structfield>classid</> column, its low-order half
> >> in the <structfield>objid</> column, and <structfield>objsubid</> equal
> >> ! to 1. The original <type>bigint</type> value can be reassembled with the
> >> ! expression <literal>(classid::int::bit(64) << 32 |
> >> ! objid::int::bit(64))::bigint</literal>. Integer keys are displayed with the
> >> ! first key in the
> >> <structfield>classid</> column, the second key in the <structfield>objid</>
> >> column, and <structfield>objsubid</> equal to 2. The actual meaning of
> >> the keys is up to the user. Advisory locks are local to each database,
>
> > Thanks, applied.
>
> This formula is not actually correct, as you'd soon find out if you
> experimented with values with the high-order bit of the low-order word
> set. (Hint: sign extension.)
>
> The correct formula is both simpler and far more efficient:
>
> (classid::int8 << 32) | objid::int8
>
> This works because oidtoi8 correctly treats the OID value as unsigned.
OK, docs updated with attached patch. Thanks. (I used 'bigint' instead
of int8 to be consistent with the surrounding text.)
--
Bruce Momjian <bruce@momjian.us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ It's impossible for everything to be true. +